SU‐FF‐J‐74: High Accuracy of Volumetric Image Registration of CT, MR and PET Images

Abstract

Purpose: To test the accuracy of the 3D volumetric image registration technique, the registration has been evaluated against “co‐registered” CT phantom images, MR/MR intramodality images and PET/CT images. Method and Materials: The 3D volumetric image registration is voxel‐based, using the homogeneous color distribution in the volumetric views of the skin voxel landmarks as the registration criterion and guidance for alignment. The software is built for up‐to‐4 concurrent imaging modality registration with real‐time volumetric manipulation and display (supported by a volume rendering board). Sixteen CT head phantom images are acquired with known spatial shifts, as well as fourteen patient cranial MR/MR (T1/T2/FLAIR) images from the same MR scanner and twenty‐five patient cranial PET/CT images from a hybrid scanner. Results: A sub‐voxel detection limit (0.1 degree/voxel) is achieved for CT/CT phantom image registration as the alignment is indicated by the color homogeneity of the aligned skin voxels, which represents a new dimension for monitoring the image registration. For the MR/MR image registration, it is found that 71% of the “co‐registered” images acquired from the same scanner within 5 minutes of each other exhibit a misalignment, caused by voluntary patient movement. The “distance” deviations between the co‐registered and voxel‐registered images are 0.2°±0.4° and 0.5±0.5 voxels. For the PET/CT image registration, 88% images have detectable misalignment due to higher probability of patient movement during longer scan time (<10 minutes) and the deviations are determined to be 0.4°±0.5° and 0.9±0.5 voxels. These movement‐induced misalignments can be corrected using the 3D volumetric image registration technique. Conclusion: The 3D volumetric image registration technique has sub‐degree/sub‐voxel accuracy in CT, MR and PET image registration. It can successfully detect misalignments in the co‐registered images visually and should be applied to correct the image misalignment caused by voluntary patient movement.